Kubeshark fixes MCP tool parameter injection vulnerability leading to cluster takeover
Kubeshark: MCP `start_kubeshark` argument injection via `pod_regex` → image redirect → cluster takeover
Kubeshark's MCP tool start_kubeshark has a parameter injection vulnerability that can lead to takeover of a Kubernetes cluster, fixed in commit a76a192. The tool concatenates the caller-supplied pod_regex as a bare positional argument into the tap command, and there is no -- option terminator, so values beginning with - are parsed by cobra as command-line flags, thereby injecting --set to override Helm values and redirect the deployed worker, hub and front images to an attacker-controlled repository; the worker DaemonSet runs privileged, with hostNetwork and the host filesystem mounted, so the attacker's image can read the host's cluster-admin kubeconfig.
The advisory states that the injection holds only when started with --allow-destructive, and includes the fix commit and permission prerequisites, so teams running the MCP service can use this to check their exposure.